Research questionHow can streaming PCA recover principal components when the data covariance changes over time?Streaming PCA processes observations sequentially, but the covariance governing those observations may change rather than remain fixed. This makes it unclear when principal-component estimates can converge and how quickly.